Skip to content

README_predict.md

Latest commit

 

History

History
83 lines (71 loc) · 2.86 KB

README_predict.md

File metadata and controls

83 lines (71 loc) · 2.86 KB
Raw

Bare Minimum Requirements to Run SignAI-SFS

Based on the codebase analysis, here are the minimum requirements to run the SignAI-SFS system:

1. Python Environment

  • Python 3.8
  • PyTorch 1.13
  • (Optional) ctcdecode 1.0.3 for more efficient beam search decoding. If not available, a custom implementation will be used.
  • Additional dependencies from requirements.txt: 
    numpy>=1.20.3
opencv-python==4.5.5.64
pandas==1.3.4
Pillow==9.4.0
PyYAML==6.0
scipy==1.7.1
six==1.16.0
tqdm==4.62.3
fvcore

2. Required Files

  • Model weights: Pre-trained model file (e.g., ./best_checkpoints/phoenix2014-T_dev_17.66_test_18.71.pt)
  • SlowFast checkpoint: ./ckpt/SLOWFAST_64x2_R101_50_50.pkl (can be downloaded with wget https://dl.fbaipublicfiles.com/pyslowfast/model_zoo/ava/pretrain/SLOWFAST_64x2_R101_50_50.pkl)
  • Configuration files:
    • ./configs/phoenix2014-T.yaml
    • ./slowfast_modules/configs/SLOWFAST_64x2_R101_50_50.yaml
  • Dictionary files:
    • ./preprocess/phoenix2014-T/gloss_dict.npy or
    • ./datasets_files/PHOENIX-2014-T/info/gloss_dict.npy

3. Input Data

  • A folder containing sign language images
  • Images should be named in a way that they are sorted correctly when using the sorted() function
    • Example: img_001.jpg, img_002.jpg, img_003.jpg, etc.

4. Hardware Requirements

  • Minimum: CPU-only system
  • Recommended: CUDA-compatible GPU
  • Also supported: Apple Silicon (M1/M2/M3) with MPS

5. Basic Command to Run

python predict.py --folder ./path/to/images --weights ./best_checkpoints/phoenix2014-T_dev_17.66_test_18.71.pt

6. Device-Specific Commands

  • For CPU-only systems:

    python predict.py --folder ./path/to/images --weights ./best_checkpoints/phoenix2014-T_dev_17.66_test_18.71.pt --device cpu

  • For systems with CUDA GPU:

    python predict.py --folder ./path/to/images --weights ./best_checkpoints/phoenix2014-T_dev_17.66_test_18.71.pt --device cuda:0

  • For Apple Silicon (M1/M2/M3) Macs:

    python predict.py --folder ./path/to/images --weights ./best_checkpoints/phoenix2014-T_dev_17.66_test_18.71.pt --device mps

7. Advanced Options

Additional parameters that can be customized:

  • --config: Path to config file (default: ./configs/phoenix2014-T.yaml)
  • --dict_path: Path to gloss dictionary
  • --search_mode: Search mode for decoding (max or beam, default: beam)
  • --input_size: Input size for the model (default: 224)
  • --image_scale: Image scale factor (default: 1.0)

8. Example Full Command with All Options

python predict.py \
  --folder ./example_images \
  --weights ./best_checkpoints/phoenix2014-T_dev_17.66_test_18.71.pt \
  --config ./configs/phoenix2014-T.yaml \
  --dict_path ./preprocess/phoenix2014-T/gloss_dict.npy \
  --device cuda:0 \
  --search_mode beam \
  --input_size 224 \
  --image_scale 1.0